In the semiconductor manufacturing process, it becomes important to immediately clear up the cause of defect generation on a semiconductor wafer in order to improve the yield. Under the existing circumstances, analysis of a defect is conducted by using a defect inspection device and a defect observation device on the spot of semiconductor manufacturing.
The defect inspection device is a device for observing a wafer by using an optical means or an electron beam and outputting location coordinates of a detected defect. In the defect inspection device, it is important to process a wide range fast. Therefore, the image data quantity is reduced by making the pixel size of an acquired image as large as possible (i.e., making the resolution low). In many cases, therefore, it is difficult to discriminate the kind of a defect (defect kind) in detail, even if existence of the defect can be confirmed from the detected image of the low resolution.
Therefore, a defect observation device is used. The defect observation device is a device for imaging coordinates of a defect on the wafer with high resolution by using output information of the defect inspection device and outputting an image. The semiconductor manufacturing process shrinks in size and consequently the defect size also reaches an order of several tens nm. For observing a defect in detail, therefore, a resolution of an order of several nm is needed.
In recent years, therefore, a defect observation device (review SEM) using a scanning electron microscope is widely used. The review SEM has an ADR (Automatic Defect Review) function of automatically collecting high resolution images of defects (defect images) on a wafer by using defect coordinates which are output by the defect inspection device.
In recent years, the throughput of the ADR of the review SEM has been improved. Therefore, it is desired to automatize the work of discriminating the defect kind on the basis of a large amount of collected defect images. The review SEM mounts an ADC (Automatic Defect Classification) function of automatically discriminating a defect kind on the basis of defect images and classifying the defect. As for the conventional art relating to the ADC, there is, for example, a technique disclosed in Patent Literature 1 described later.